Time and delays

The kernel keeps one monotonic clock, the tick count, advanced by the tick interrupt and by pbl_idle_slept() after tickless sleep. pbl/kernel/time.h reads it and the CPU cycle counter; pbl/kernel/delay.h waits on them.

Uptime

pbl_uptime_ticks() and pbl_uptime_ms() count from the scheduler start in 64 bits, so they do not wrap during the life of a device. The 32-bit pbl_tick_t remains for relative timeouts only.

pbl_cycle_get_32() reads the CPU cycle counter (DWT CYCCNT on Cortex-M) at pbl_cycles_per_sec(), the current CPU clock. Use it to time short sections: it wraps within seconds and stops while the CPU sleeps.

Waiting

Need	Use	Resolution	Context
A few microseconds, or no thread	`pbl_busy_wait_us()`	cycles	anywhere
Microseconds, being late is fine	`pbl_sleep_us()`	one tick	thread
Milliseconds and longer	`pbl_sleep_ms()`	one tick	thread
A timeout or a deadline in ticks	`pbl_thread_sleep()`	one tick	thread
A condition, spinning	`PBL_POLL_BUSY_US()`	cycles	anywhere
A condition, sleeping	`PBL_POLL_SLEEP_MS()`	one tick	thread

The pbl_sleep_*() and pbl_busy_wait_us() calls wait at least the requested time, which is what hardware timings from a datasheet need: sleeps round up to whole ticks and add one for the tick already in progress. pbl_thread_sleep() counts tick boundaries instead, so its first tick may be partial.

The polling macros return 0 once the condition holds and -ETIMEDOUT otherwise, evaluating it once more after the timeout so that preemption cannot turn a ready condition into a false timeout, as Linux read_poll_timeout() does.

Busy waiting keeps the CPU awake and starves lower priority threads; past a tick or so, sleep instead.

PebbleOS

Time and delays

Uptime

Waiting

Overview